Memory is expensive, and most of the memory that is required to process encoded digital video data streams is used to store luminance and chrominance values for each pixel in a frame of data. Information for a frame must be stored in memory while that frame is decoded and processed. Although reducing memory requirements is attractive for any application, it is especially important for applications in hand-held and low-power devices like mobile telephones and so called personal digital assistants (PDAs) because limited circuit board space and power severely restrict memory chip size and power requirements.
Many applications conform to one of two coding standards that reduce memory requirements by using data compression techniques. The H.263 standard, published by the International Telecommunications Union (ITU), supports video compression for video-conferencing and video-telephony applications. The MPEG-4 standard, known officially as ISO/IEC 14496, is published by the International Organization for Standardization (ISO) and facilitates the integration of digital television, interactive graphics and interactive multimedia. These techniques can reduce the memory required by processes that transmit or store the information, but they do not address the amount of memory required by processes that decode the encoded information.
Attempts to provide additional compression encounter several problems. One problem is caused by many compression schemes that introduce small errors into the digital video data and the magnitude of these errors accumulates during decoding. Small inaccuracies caused by rounding or truncation, for example, are amplified in predictive filters used by the decoding process because the errors in each stage cause larger errors in a later stage of prediction.
Another problem is caused by the fact individual pixels throughout an image must be accessed quickly for processes like motion compensation. This requirement makes compression techniques that rely on variable-length symbols unattractive because random accesses into memory are more difficult.